The dynamic spatial multinomial probit model: analysis of land use change using parcel-level data

Abstract

Many transportation-related behaviors involve multinomial discrete response in a temporal and spatial context. These include quality of paved roadway sections over time, evolution of land use at the parcel level, vehicle purchases by socially networked households, and mode choices by individuals residing across adjacent homes or neighborhoods. Such responses depend on various influential factors, and can have temporal and spatial dependence or autocorrelation. In many cases, dynamic spatial-model specifications based on maximum fitness, profit or utility may be most appropriate.This study develops a Dynamic Spatial Multinomial Probit (DSMNP) model by pivoting off the ordinary MNP model while incorporating spatial and temporal dependencies. The paper first explains how the model reflects the behaviors at play, and then describes estimation using Bayesian methods. Simulated data sets containing both generic and alternative-specific explanatory variables are used to validate the model's performance (and that of its associated code). Estimation efficiency issues and identification issues are discussed. The model is then applied to analyze parcel-level land use changes in Austin, Texas. It is found that better accessibility boosts the potential of residential development, with a statistically significant parameter. In the case of non-residential development, the effect is statistically insignificant. The effects of job and population density, neighborhood income and soil slope are also explored, and found to exert variable effects across space. It is also found that land development tends to cluster when existing development intensity in a neighborhood is low.2

Spatio-temporal multinomial autologistic modeling of land-use change: A parcel-level approach

“…They use a sample of 783 parcels in Austin over the years 2000, 2003, 2006, and 2008, and consider four mutually exclusive land developments of increasing intensity: no development, single and two-family residential, other residential, and nonresidential. Finally, Wang et al (2012) present a multinomial probit model that incorporates spatial and temporal dependencies to investigate parcel-level land-use changes in Austin over 8 years (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007). Both dependencies are related to the error term, and not to the latent utility variable.…”

A spatial autoregressive multinomial probit model for anticipating land-use change in Austin, Texas

“…One expects this, since isolated developments are not as meaningful to their buyers/users as are coordinated developments, with greater neighborhood interaction, due to lowered travel costs and higher agglomeration benefits Marginal effects are computed as the averaged changes in probabilities (by land use types) per one standard deviation (SD) increase in one of the covariates, holding everything else constant. The SAR MNP model allows one to analyze direct and indirect spatial effects separately, as opposed to the models developed by Chakir and Parent (2009) and Wang et al (2011), which yield marginal effects indistinguishable from those of the standard MNP model (since they assume that spatial autocorrelation occurs only across the error terms, rather than across latent response terms). Unlike LeSage and Pace's (2009) bivariate spatial example, the SAR MNP model's marginal effects (for a particular land use type k and covariate p) are represented as a series of • n x n matrices (where n is the total number of observations/locations, or 1,500 in this paper), with the ij th cell value indicating changes in land use k's probability at location i following a one-SD increase in the p th covariate's value at location j (K is the total number of land use alternatives and P the total number of covariates minus the constant term).…”

Validating activity, time, and space diversity as essential components of urban vitality

“…Tan and Wu (2003) adopted information entropy values in the regulation of land use structure. Those previous studies of mixed land use heavily relied on parcel-level land use maps (Wang et al., 2012). Additionally, recent studies suggested the feasibility of using POIs (Abdullahi et al., 2015; Jiang et al., 2015; Liu and Long, 2016; Yao et al., 2017) as an alternative to land use data (Wang et al., 2012), in that POIs can represent a much finer grained picture of land use at the building level (Kunze and Hecht, 2015).…”